

Ci a ion: Alonso de A miño, C.;
U da, D.; Alcalde, R.; Ga cía, S.;
He e o, Á. An In elligen
Visualisa ion Tool o Analyse he
Sus ainabili y o Road T anspo a ion.
Sus ainabili y 2022,14, 777. h ps://
doi.o g/10.3390/su14020777
Academic Edi o : Ma c A. Rosen
Recei ed: 7 Oc obe 2021
Accep ed: 7 Janua y 2022
Published: 11 Janua y 2022
Publishe ’s No e: MDPI s ays neu al
wi h ega d o ju isdic ional claims in
published maps and ins i u ional a il-
ia ions.
Copy igh : © 2022 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license (h ps://
c ea i ecommons.o g/licenses/by/
4.0/).
sus ainabili y
A icle
An In elligen Visualisa ion Tool o Analyse he Sus ainabili y
o Road T anspo a ion
Ca los Alonso de A miño 1, Daniel U da 2,* , Robe o Alcalde 3, San iago Ga cía1and Ál a o He e o 2
1Depa amen o de Ingenie ía de O ganización, Escuela Poli écnica Supe io , Uni e sidad de Bu gos,
A . Can ab ia S/N, 09006 Bu gos, Spain; [email p o ec ed] (C.A.d.A.); [email p o ec ed] (S.G.)
2G upo de In eligencia Compu acional Aplicada-GICAP, Depa amen o de Ingenie ía In o má ica,
Escuela Poli écnica Supe io , Uni e sidad de Bu gos, A . Can ab ia S/N, 09006 Bu gos, Spain;
[email p o ec ed]
3Depa amen o de Economía y Adminis ación de Emp esas, Facul ad de Ciencias Económicas y
Emp esa iales, Uni e sidad de Bu gos, Pza. de la In an a Dª. Elena, S/N, 09001 Bu gos, Spain;
[email p o ec ed]
*Co espondence: du [email p o ec ed]
Abs ac :
Road anspo is an in eg al pa o economic ac i i y and is he e o e essen ial o i s
de elopmen . On he downside, i accoun s o 30% o he wo ld’s GHG emissions, almos a hi d
o which co espond o he anspo o eigh in hea y goods ehicles by oad. Addi ionally,
means o anspo a e s ill e ol ing echnically and a e subjec o e e mo e demanding egula ions,
which aim o educe hei emissions. In o de o analyse he sus ainabili y o his ac i i y, his
s udy p oposes he applica ion o no el A i icial In elligence echniques (mo e speci ically, Machine
Lea ning). In his esea ch, he use o Hyb id Unsupe ised Explo a o y Plo s is b oadened wi h
new Explo a o y P ojec ion Pu sui echniques. These, oge he wi h clus e ing echniques, o m
an in elligen isualisa ion ool ha allows knowledge o be ob ained om a p e iously unknown
da ase . The p oposal is es ed wi h a la ge da ase om he o icial su ey o oad anspo in Spain,
which was conduc ed o e a pe iod o 7 yea s. The esul s ob ained a e in e es ing and p o ide
encou aging e idence o he use o his ool as a means o in elligen analysis on he subjec o
de elopmen s in he sus ainabili y o oad anspo a ion.
Keywo ds:
a i icial in elligence; unsupe ised machine lea ning; explo a o y p ojec ion pu sui ;
clus e ing; oad anspo a ion; anspo sus ainabili y; age o anspo means
1. In oduc ion & P e ious Wo k
The 17 Sus ainable De elopmen Goals o he UN’s mas e plan include a de e mined
line o ac ion o hal global wa ming, wi h a di ec link o educing CO
2
emissions. The
Uni ed Na ions F amewo k Con en ion on Clima e Change (UNFCCC) and he Kyo o
P o ocol, bo h signed in 2002, poin in he same di ec ion.
Mo eo e , oad anspo is consubs an ial o economic ac i i y and, hus, essen ial o
he u u e o ou ci ilisa ion. I s coun e pa accoun ed o 30% [
1
] o global CO
2
emissions
in 2015, and 29% o he 314,529 kilo- onnes o GHG emi ed in 2019 in Spain, wi h a o al o
91,372 kilo- onnes o CO
2
-equi alen emissions. In u n, i is es ima ed a a global and an
EU le el ha nea ly a qua e o hese emissions co espond o eigh anspo a ion in
hea y goods ehicles and buses [
2
,
3
], wi h his being one o he a eas ha has eco ded
sus ained g ow h in ecen yea s. Meanwhile, o he modes o anspo a e p og essi ely
educing hei sha e, wi h an es ima ed inc ease o ou imes hei cu en emissions by
2050 i no measu es a e adop ed on he ma e . In iew o hese da a and o ecas s, pa
o he ocus o scien i ic s udies and lines o ac ion o go e nance ha e cen ed on his
a ea, wi h p og ess being made on he de elopmen o p oposals o egula e emissions o
he equipmen ha pe o ms his ac i i y, which clea ly accoun s o 6% o he Eu opean
Union’s o al GHG emissions [4].
Sus ainabili y 2022,14, 777. h ps://doi.o g/10.3390/su14020777 h ps://www.mdpi.com/jou nal/sus ainabili y
Sus ainabili y 2022,14, 777 2 o 15
1.1. Sus ainabili y in T anspo a ion
The sys ema ic e iew o in e en ion sys ems o he sus ainabili y o oad eigh
anspo e ealed ha he mos ob ious line o ac ion is he applica ion o eme ging ech-
nologies combined wi h decisi e ene gy policies [
5
]. O he complemen a y lines o ac ion
ocus on he dedica ion and adap a ion o in as uc u es o he e ec i e de elopmen o
his ac i i y and he p omo ion o in e modali y in he e icien managemen o supply
chains [
6
]. In many cases, hese a e cen ed on a i m commi men o synch o-modali y [
7
]
and a e e en inclined o measu e he sus ainabili y o anspo on he basis o i s in e -
modali y [
8
]. A dis inc ion is also made be ween he e ec i eness o di e en measu es in
a ou o sus ainabili y in collec ion and deli e y anspo , wi h imp o ed applicabili y o
collabo a i e echniques suppo ed by in o ma ion sys ems, as opposed o long-dis ance
anspo , whe e he esul s o echnological op imisa ion in e ms o anspo equipmen
consump ion and he sui abili y o in as uc u es a e mo e con incing [
9
]. Some models,
e lec ing he close connec ion be ween he economy and anspo a ion, a e based on he
analysis o an economic sus ainabili y app oach o logis ics models [
10
] and, along he
same lines, some o hem ocus on he analysis o ce ain goods based on hei p oduc ion
s uc u e and he dis ibu ion o goods by oad, analysing hei en i e ac i i y [
11
]. The e
a e also s udies aimed a op imizing he anspo p ocess based on s eamlining some o
i s sub-p ocesses such as loading [12,13].
To his ex en , we could say ha we a e ollowing a classic pe spec i e in he sea ch o
sus ainabili y, which Co lu [
14
] poin s o in his e iew o he s a e o he a on op imising
ene gy consump ion in anspo p ocesses, as a combina ion o h ee main lines o ac ion
o oad anspo : (i) imp o ing he load ac o , unde s ood as a educe o emissions
based on maximum occupancy o he means o anspo , (ii) he use o collabo a i e
echniques o op imise he alloca ion o means o anspo and (iii) de ining and moni o ing
sus ainabili y objec i es in he de elopmen o anspo ope a ions. Ne e heless, he
mos ema kable aspec o he ensemble is ha all he models and s udies poin , in sho ,
o a common unde lying objec i e; he op imisa ion o ene gy consumed in he p ocess o
eigh anspo ope a ions.
A line o esea ch di ec ly associa ed wi h he cen al objec i e o ene gy imp o emen
is he s udy o emissions om anspo equipmen based on hei age. Hassani e al. [
15
]
de e mined ha emissions in ligh ehicles can be up o i e imes highe depending on
hei age. Owing o his inding, his idea (sus ainabili y based on he ehicles age) is gi en
special ele ance in his s udy.
These esul s a e no a bi a y; ins ead, hey a e he esul o he implemen a ion
o policies adop ed by he ehicle manu ac u ing sec o . Back in 2002, Ang-Olson and
Sch oee [
16
] de e mined ha he p ope implemen a ion o echnical solu ions in he
p oduc ion o hea y goods oad anspo ehicles could lead o a educ ion o mo e han
11 billion li es o annual uel consump ion wi hin a pe iod o 10 yea s, associa ed wi h
a dec ease o 8.3 million onnes o g eenhouse gas emissions in he Uni ed S a es (US)
alone. Since hen, he go e nance mechanisms o he US and he EU ha e con inued o
impose manda es on he p oduc ion o anspo ehicles, aimed a educing consump ion
and emissions [
17
]. As a esul o hese manda es, he so-called Eu o emission s anda ds
ha e been p og essi ely de eloped, wi h successi ely mo e demanding equi emen s on
consump ion and emissions o anspo elemen s, as shown in Table 1, on ca bon monoxide
(CO), ni ogen oxides (NOX) and pa icula e ma e (PM).
I we ocus speci ically on hea y goods ehicles, Haugen and Bishop [
19
] es ablish
wo comple e ehicle emission samplings a loading and unloading poin s wi h signi ican
eigh mo emen s, inally de e mining ha a educ ion om an a e age age o 7.8 o
6 yea s o hese ehicles esul s in a signi ican educ ion in emissions o up o 87% o
suspended pa icula e ma e . A subsequen s udy [
20
] also de e mined ha he educ ion
in ha m ul NOX emissions dec eased om 38 g o 9 g pe kg o diesel uel consumed
by hea y goods ehicles om 2005 o 2020, ep esen ing a educ ion o 76.3% in ehicles
Sus ainabili y 2022,14, 777 3 o 15
p oduced be ween hese da es, i.e., an a e age educ ion o 5% in emissions o each yea
o ehicle p oduc ion du ing his pe iod.
Table 1.
Emission s anda ds adop ed by he EU o diesel ca s and hea y goods ehicles. Sou ce: Own
elabo a ion on he da a o [18].
Emission S anda ds o Diesel Ca s
S anda d Da e CO g/Km NOX g/Km PM g/Km
Eu o 4 2005 0.50 0.30 0.025
Eu o 5 2010 0.50 0.23 0.005
Eu o 6 2015 0.50 0.17 0.005
Emission S anda ds o Hea y Goods Vehicles
S anda d Da e CO g/KWh NOX g/KWh PM g/KWh
Eu o IV 2005 1.50 3.50 0.020
Eu o V 2008 1.50 2.00 0.020
Eu o VI 2013 1.50 0.40 0.010
Wi h possible di e ences in he quan i ica ion o emission imp o emen s, he e is
one clea conclusion; hea y goods ehicles a e a signi ican con ibu o o he emission o
ha m ul gases, and he age o said ehicles is also a de e mining ac o in hei e iciency
and sus ainabili y.
1.2. P e ious Wo k on Digi isa ion
New pe spec i es eme ge when inco po a ing he digi alisa ion app oach o oad
anspo a ion. A s udy by a panel o 52 expe s [
21
], highligh s he alue o p ocess
au oma ion, he comple e collec ion o da a on digi al in o ma ion and he basis o he
applica ion o a i icial in elligence o adequa e pe o mance and planning. Taking a s ep
u he in his di ec ion, and wi h a clea ocus on imp o ing he managemen o supply
chains, a s udy was ca ied ou on he da a collec ed by he EU in he Pe manen Su eys
o Goods T anspo by Road [
22
]. Wi h his pu pose, a modelling was ca ied ou on his
ac i i y, which was compiled by EUROSTAT o i s membe coun ies be ween 2011 and
2014. I was ca ied ou unde a Ho izon al Collabo a ion model [
23
] and analysed he
imp o emen s ha would ha e esul ed om he applica ion o Pooling echniques, simila
o op imised eigh g oupings, and an implemen a ion model o he so-called Physical
In e ne [
24
]. This was done by de eloping compu e models ha simula e each op ion,
which a e used o es ima e a calcula ion o emissions. The end esul is a clea ad an age
o he cu en ly pu ely heo e ical physical in e ne model.
Mo e speci ically, p e ious con ibu ions ha e been made conce ning he use o A i i-
cial In elligence (AI) in gene al and Machine Lea ning (ML) in pa icula , in o de o add ess
sus ainabili y issues in oad eigh anspo . In [
25
], an in eg a ed uzzy ailu e mode
and e ec s analysis app oach was p oposed o he selec ion o isk mi iga ion s a egies
in ack and ace asks in he indus y, which aimed o help manage s choose a s a egy
conside ing he c i icali y o he isks unde a limi ed budge . In addi ion, [
26
] p esen ed a
no el con aine anspo op imisa ion model ha inco po a es he oad ne wo k along
wi h connec i i y me ics, aiming o minimise o al ip dis ance, uck uel cos , con aine
en al cos , and con aine mo emen s be ween mul iple consignees and haule s. Mo e
ecen ly, [
27
] p oposed a no el app oach o p edic he p o i ma gin, on a cus ome basis,
in he sus ainable oad eigh anspo sec o by combining di e en ML me hods. This
helps manage s o ob ain use ul in o ma ion on s a egic and sus ainable de elopmen
pe spec i es.
Di e ing om his p e ious s udy, he au ho s o his a icle ha e applied di e en
ML echniques [
28
] o he same da a amily (see Sec ion 2). Signi ican esul s ha e been
ob ained wi h espec o hei link wi h he economic ac i i y cycles. The economic u ning
poin s ha occu ed du ing he G ea Recession in Spain ha e also been iden i ied, based
exclusi ely on a ious clus e ing echniques om he oad eigh anspo da ase . A
Sus ainabili y 2022,14, 777 4 o 15
s ong ecession was obse ed un il he second qua e o 2012, ollowed by a se e e
dep ession du ing he ollowing pe iod un il he second qua e o 2013, which was hen
ollowed by a g adual eco e y un il in he second qua e o 2015, whe e a clea phase o
economic g ow h eme ged.
The da ase showing he a e age lee age o hea y goods oad ehicles ne e ceased
o inc ease a any gi en momen du ing he pe iod in ques ion (al hough i inc eased a
di e en a es). I almos exac ly coincided a he poin s o economic in lec ion wi h he
qua ile (Q) dis ibu ion o he da a, as can be seen in Figu e 1. This accele a ed o sus ained
g ow h co esponds o no hing o he han a pa e n o o e -amo isa ion o he means o
p oduc ion in imes o economic ecession, as a measu e o p o ec he p o i abili y o i s
economic ac i i y.
Sus ainabili y2022,14,xFORPEERREVIEW4o 15

me hods.Thishelpsmanage s oob ainuse ulin o ma ionons a egicandsus ainable
de elopmen pe spec i es.
Di e ing om hisp e iouss udy, heau ho so  hisa icleha eapplieddi e en 
ML echniques[28] o hesameda a amily(seeSec ion2).Signi ican  esul sha ebeen
ob ainedwi h espec  o hei linkwi h heeconomicac i i ycycles.Theeconomic u ning
poin s ha occu eddu ing heG ea RecessioninSpainha ealsobeeniden i ied,based
exclusi elyon a iousclus e ing echniques om he oad eigh  anspo da ase .A
s ong ecessionwasobse edun il hesecondqua e o 2012, ollowedbyase e ede‐
p essiondu ing he ollowingpe iodun il hesecondqua e o 2013,whichwas hen
ollowedbyag adual eco e yun ilin hesecondqua e o 2015,whe eaclea phaseo 
economicg ow heme ged.
Theda ase showing hea e age lee ageo hea ygoods oad ehiclesne e ceased
oinc easea anygi enmomen du ing hepe iodinques ion(al houghi inc easeda 
di e en  a es).I almos exac lycoincideda  hepoin so economicin lec ionwi h he
qua ile(Q)dis ibu iono  heda a,ascanbeseeninFigu e1.Thisaccele a edo sus‐
ainedg ow hco esponds ono hingo he  hanapa e no o e ‐amo isa iono  he
meanso p oduc ionin imeso economic ecession,asameasu e op o ec  hep o i a‐
bili yo i seconomicac i i y.

Figu e1.T anspo  lee ageda ase ies,di idedin oqua ilesand ela ed o hephasesde e mined
o  heG ea Dep ession.Sou ce:Ownelabo a ion.
Weha edecided o ocusou s udieson hisda ase iesg oupedin oqua ilesowing
o he ollowing easons:Fi s ly,i isspeci icallylinked o hesus ainabili yo  anspo 
ac i i y;andsecondly,i sdis ibu iono da abyqua ilescoincideswi h heeconomic
phases.Fu he mo e, hes udyo  hisse ies,whichiskey o hein e p e a iono  hesus‐
ainabili yo  he lee o  eigh  anspo  ehicles,wasneglec ed om hee olu ionand
s udyo  he es o  hese ies ha  e lec edspeci icinc easeso dec easesin hedep ession
phase,whichmakesi ad isable oapplycomplemen a yanalysis echniques oi .
Toadd ess hisp oblem,pionee  isualisa ion oolsbasedonMLa epu  o wa din
hiss udy.Mo especi ically,clus e ingandExplo a o yP ojec ionPu sui (EPP) ech‐
niquesha ebeencombined o  he i s  ime,unde  he ameo Hyb idUnsupe ised
Explo a o yPlo s(HUEPs), osuppo  he isualanalysiso sus ainabili yda a ega ding
oad anspo a ion.Likewise, heo iginal o mula iono HUEPsisex ended,aswellas
newp ojec ion echniquesbeingp oposedand alida ed.
Thus a , esea che sha ewidelys udiedclus e ingandEPPme hods,concluding
someo  hem ha p ojec ionme hodsa eno ause ulino de  o educe hedimension‐
ali yo da a o a ollowingclus e ing[29,30].Al hough hiss a emen maybe uein
Figu e 1.
T anspo lee age da a se ies, di ided in o qua iles and ela ed o he phases de e mined
o he G ea Dep ession. Sou ce: Own elabo a ion.
We ha e decided o ocus ou s udies on his da a se ies g ouped in o qua iles owing
o he ollowing easons: Fi s ly, i is speci ically linked o he sus ainabili y o anspo
ac i i y; and secondly, i s dis ibu ion o da a by qua iles coincides wi h he economic
phases. Fu he mo e, he s udy o his se ies, which is key o he in e p e a ion o he
sus ainabili y o he lee o eigh anspo ehicles, was neglec ed om he e olu ion and
s udy o he es o he se ies ha e lec ed speci ic inc eases o dec eases in he dep ession
phase, which makes i ad isable o apply complemen a y analysis echniques o i .
To add ess his p oblem, pionee isualisa ion ools based on ML a e pu o wa d
in his s udy. Mo e speci ically, clus e ing and Explo a o y P ojec ion Pu sui (EPP) ech-
niques ha e been combined o he i s ime, unde he ame o Hyb id Unsupe ised
Explo a o y Plo s (HUEPs), o suppo he isual analysis o sus ainabili y da a ega ding
oad anspo a ion. Likewise, he o iginal o mula ion o HUEPs is ex ended, as well as
new p ojec ion echniques being p oposed and alida ed.
Thus a , esea che s ha e widely s udied clus e ing and EPP me hods, concluding
some o hem ha p ojec ion me hods a e no a use ul in o de o educe he dimensionali y
o da a o a ollowing clus e ing [
29
,
30
]. Al hough his s a emen may be ue in some cases,
some o he combina ions o such me hods ha e been p e iously p oposed, di e en om
his sequen ial applica ion o me hods. Tha is he case o [
31
,
32
], whe e he ou pu o clus-
e ing me hods (i.e., he assigned clus e o each da a ins ance) is added o he p ojec ion
ob ained by EPP me hods, ha could be 2D o 3D. On he o he hand, o he au ho s ha e
p oposed [
33
,
34
] he simul aneous applica ion o clus e ing and dimensionali y- educ ion
me hods. As opposed o hese p e ious ideas, HUEPs ha e been ecen ly p oposed o he
combina ion o clus e ing and p ojec ion me hods, being independen ly applied.
Sus ainabili y 2022,14, 777 5 o 15
The emaining sec ions o his a icle a e o ganised as ollows: he me hods employed,
oge he wi h he da a on which hey a e applied, a e desc ibed in Sec ion 2. The esul s
ob ained in he expe imen al s udy a e p esen ed in Sec ion 3, and he main conclusions in
ela ion o hese, as well as some p oposals o u u e wo k, a e p esen ed in Sec ion 4.
2. Ma e ials and Me hods
As p e iously s a ed, oad anspo a ion ac i i y was esea ched in his s udy, wi h
a speci ic ocus on i s sus ainabili y. This was done by analysing a da ase desc ibed in
Sec ion 2.1 wi h he no el echniques ha a e p esen ed in Sec ion 2.2.
2.1. Da ase
Da a we e e ie ed om wo di e en sou ces:
•
The Minis y o T anspo , Mobili y and U ban Agenda (Minis e io de T anspo es,
Mo ilidad y Agenda U bana) o Spain, h ough i s Gene al Sub-Di ec o a e o Economic
S udies and S a is ics.
•
The Eu opean Road F eigh T anspo su ey (ERFT). This su ey ela es o he ac i i y
o hea y goods ehicles licenced in Spain o he anspo o goods. I has a su icien ly
high sampling le el o be o s a is ical ep esen a i eness o each Au onomous Region,
in o de o measu e hei anspo ope a ions. Wi h his aim, he su ey egis e s
he mo emen o a single class o goods, om a depa u e poin o a des ina ion. The
esea ch was conduc ed in acco dance wi h he co esponding egula ion [
35
] and i s
subsequen e ision [
36
]. The o al numbe o eco ds included on ha da abase was
1,932,671 ha has a sampling ep esen a i eness o 1,259,938,252 anspo ope a ions.
Da a om be ween 2011 and 2017 we e used. All he da a ep esen ed qua e ly le els
o agg ega ion, which he e o e included each a iable in he s udy, a o al o 28 alues.
The a iables unde conside a ion we e:
•
T anspo a ion cos s (B): based on a 100 pe cen inc ease abo e he a e age yea ly
p ices in 2000, as de e mined by he Minis y o De elopmen ’s qua e ly esea ch
s udies.
•Fuel cos s in Spain: qua e ly midpoin s weigh ed in cen imes o a eu o, as indica ed
by he da a ga he ed by he Minis y o De elopmen .
•
Fuel cos s in he EU: qua e ly midpoin s weigh ed in cen imes o a eu o, as indica ed
by he da a ga he ed by he Minis y o De elopmen .
•Numbe o ons anspo ed (A, B, C): weigh o anspo ed goods.
•Comple ed ips (A, B, C): numbe o anspo ope a ions and emp y dis ance.
•Emp y dis ance (A, B): kilome es a elled wi hou goods.
•
Maximum load o anspo ope a ions (A, B): uppe weigh limi o comple ed ips
in ons.
•
Maximum load o emp y dis ance (A, B): uppe weigh limi o emp y dis ance
co e ed in ons.
•Haulage dis ance (A, B): kilome es a elled.
•Emp y haulage dis ance (A, B): kilome es a elled wi hou goods.
•Quan i y o ehicles ep esen ed (A): numbe o ehicles ep esen ed.
•Rep esen ed load capaci y (A): uppe load limi o he ep esen ed ehicles.
•
Tons-kms (A, B, C): o al ons anspo ed, and dis ance co e ed in each haulage
ope a ion.
•
A e age lee age (A, B): a e age amoun o yea s elapsed since he egis a ion o he
ehicles. As p e iously indica ed in Sec ion 1.1, his is an impo an da a inding in
ega d o sus ainabili y. Owing o his, i is also used in he glyph me apho .
•
A e age lee age o emp y dis ance (A, B): a e age amoun o yea s elapsed since he
egis a ion o he ehicles a elling wi hou goods.

Sus ainabili y 2022,14, 777 6 o 15
The da a se ies wi h assigned le e s we e sub-di ided as acco ding:
(A)
Type o anspo : (A1) All anspo ; (A2) Own anspo ; (A3) Hi e o ewa d.
(B)
Dis ance ange: (B1) All dis ances; (B2) < 50 km; (B3) 51–100 km; (B4) 101–200 km;
(B5) 201–300 km; (B6) > 300 km.
(C)
Geog aphic ca chmen : (C1) All ca chmen s; (C2) Municipal; (C3) Regional; (C4)
Na ional; (C5) Impo a ion; (C6) Expo a ion; (C7) Cabo age.
113 anspo da a se ies we e calcula ed, wi h he alues o he 28 p e iously indica ed
qua e s in each one.
As a esul , a da ase wi h high dimensionali y is equi ed o be analysed in o de o
in es iga e he sus ainabili y o oad anspo a ion.
2.2. Hyb id Unsupe ised Explo a o y Plo s
Hyb id Unsupe ised Explo a o y Plo s (HUEPs) [
37
] ha e been ecen ly p oposed as
a new isualisa ion ool o combine he ou pu s o Explo a o y P ojec ion Pu sui (EPP)
and clus e ing me hods in a no el and in o ma i e way. To add ess he well-known “cu se
o dimensionali y” challenge and ad ancing in desc ip i e da a analysis, bo h EPP and
clus e ing me hods a e independen ly applied, and hei ou pu s combined in a new way.
In pa icula , 3 EPP me hods we e pu o wa d, commonly known as P incipal Compo-
nen Analysis (PCA), Maximum Likelihood Hebbian Lea ning (MLHL), and Coope a i e
MLHL (CMLHL). The e a e di e en ways o implemen ing such me hods; in he o iginal
o mula ion o HUEPs, hey we e implemen ed as A i icial Neu al Ne wo ks.
Addi ionally, an ex ension o his s udy is included o imp o e he isualisa ion
capabili y o HUEPs. In o de o gene a e he displays, each o iginal x ec o ( om he
inpu space) is p ocessed as ollows:
1. 2D p ojec ion o he ec o is ob ained by he applied EPP me hod (yEPP
1,yEPP
2).
2.
The ou pu o he clus e ing me hod (i.e., he assigned clus e numbe ) is calcula ed
(yc).
3.
The wo p e ious ou pu s a e combined in a 3D ec o ha is loca ed in he ou pu
space (y1,y2,y3).
4.
Op ionally, u he in o ma ion (sus ainabili y da a in he p esen s udy) is added o
he isualisa ion by using he glyph me apho .
O iginally, HUEPs we e concei ed as a new way o in ui i ely displaying da a by
applying one pa i ional (k-means) o one hie a chical (agglome a i e) clus e ing me hod
oge he wi h one EPP me hod. As an e olu ion o his ini ial p oposal, his s udy alida es
he inco po a ion o complemen a y and well-known display me hods, namely Ke nel-PCA
(KPCA) [38] and Sammon Mapping (SM) [39].
KPCA is a non-linea ex ension o con en ional PCA ha akes he majo i y o ke nel
unc ions in o de o ob ain mo e in e es ing p ojec ions o da a by ex ac ing non-lineal
p incipal componen s while keeping he compu a ion cos a a easonable le el. On he
o he hand, SM was p oposed as a special case o he dis ance-based me ic Mul idimen-
sional Scaling amily, being i sel one o he i s mani old lea ning p oposals. Fu he mo e,
SM can be conside ed as he i s p oposed nonlinea mani old lea ning me hod. These
non-linea EPP me hods a e p oposed o he i s ime unde he ame o HUEPs as being
ones o he main non-linea EPP me hods. They a e analysed in his s udy and alida ed
wi h he da a p e iously desc ibed.
2.3. Meh odology
In o de o alida e he p oposed applica ion o HUEPs in he p esen wo k, isual-
iza ions ha e been ob ained by combining he p ojec ions o EPP me hods (PCA, MLHL,
CMLHL, KPCA, and SM) wi h he ou pu o clus e ing me hods (k-means and agglome a-
i e). Expe imen s ha e been pe o med uning each one o he me hods wi h he ollowing
pa ame e alues.
Sus ainabili y 2022,14, 777 7 o 15
PCA
•Numbe o p incipal componen s o be ob ained: 2.
MLHL
•Numbe o p ojec ed dimensions o be ob ained: 2.
•Lea ning a e: [0.01, 0.05].
•ppa ame e : [1, 2].
CMLHL
•Numbe o p ojec ed dimensions o be ob ained: 2.
•Lea ning a e: [0.01, 0.05].
•ppa ame e : [1, 2].
• au pa ame e : [0.00000001, 0.01].
KPCA
•Numbe o p ojec ed dimensions o be ob ained: 2, 3.
•Ke nel: linea , polynomial, gaussian.
SM
•Numbe o p ojec ed dimensions o be ob ained: 2, 3.
•k-means.
•Numbe o clus e s: 2, 3, 4, 6, 8.
•Dis ance: sqEuclidean, Ci yblock, Cosine, Co ela ion.
Agglome a i e
•Numbe o clus e s (cu o ): 2, 3, 4, 6, 8.
•
Dis ance: Euclidean, sEuclidean, sqEuclidean, Ci yblock, Hamming, Jacca d, Minkowski,
Chebyche , Spea man, Cosine, Co ela ion.
•Linkage: A e age, Cen oid, Comple e, Median, Single, Wa d, Weigh ed.
3. Resul s
The HUEP displays ob ained a e shown in his sec ion. Fi s ly, Figu e 2shows he
HUEP display ob ained by combining agglome a i e clus e ing wi h di e en EPP ech-
niques. As a esul , he in luence o he di e en EPP echniques on he ob ained esul s
can be compa ed. Due o his, addi ional in o ma ion is no shown h ough he glyph
me apho in his igu e, o enable he sole compa ison o he p ojec ions. Fo he sake o
b e i y, he mos in e es ing g aphical displays a e shown and hose ob ained by some o
he EPP me hods a e no included in Figu e 2.
The display ob ained by KPCA can be conside ed he mos e ealing. I allows he
s uc u e o he da a o be obse ed mo e clea ly, as i ep esen s he da a in a mo e compac
o m and hus allows ends o be analysed wi h mo e cla i y. Since i is no possible o
include all he esul s ob ained in his s udy, only esul s ob ained using KPCA a e shown
in he es o his sec ion.
These esul s alida e he main p oposal o he p esen esea ch: ex ending he o iginal
HUEP o mula ion by adding new EPP me hods ha can imp o e he isualiza ion o a
gi en da ase . Fo he da ase unde analysis, none o he EPP me hods in he o iginal
HUEP o mula ion p o ides wi h he bes p ojec ion, bu one o he new ones (KPCA)
ins ead.
Sus ainabili y 2022,14, 777 8 o 15
Sus ainabili y2022,14,xFORPEERREVIEW8o 15


(a)

(b)

(c)
Figu e2.HUEPsob ainedbyapplyingagglome a i eclus e ing(k=3,dis ance=sEuclidean,link‐
age=a e age) o heanalysedda ase , a ying heEPP echnique:(a)CMLHL,(b)KPCA,(c)SM.
Thedisplayob ainedbyKPCAcanbeconside ed hemos  e ealing.I allows he
s uc u eo  heda a obeobse edmo eclea ly,asi  ep esen s heda ainamo ecom‐
pac  o mand husallows ends obeanalysedwi hmo ecla i y.Sincei isno possible
oincludeall he esul sob ainedin hiss udy,only esul sob ainedusingKPCAa e
shownin he es o  hissec ion.
These esul s alida e hemainp oposalo  hep esen  esea ch:ex ending heo ig‐
inalHUEP o mula ionbyaddingnewEPPme hods ha canimp o e he isualiza ion
o agi enda ase .Fo  heda ase unde analysis,noneo  heEPPme hodsin heo iginal
Figu e 2.
HUEPs ob ained by applying agglome a i e clus e ing (k = 3, dis ance = sEuclidean,
linkage = a e age) o he analysed da ase , a ying he EPP echnique: (
a
) CMLHL, (
b
) KPCA, (
c
) SM.
Sus ainabili y 2022,14, 777 9 o 15
Resul s including he Glyph Me apho
A e ha ing selec ed KPCA as he EPP model ha o e s he bes p ojec ions o he
da a analysed, he esul s using he glyph me apho a e p esen ed in his sec ion. I is
wo h men ioning ha o any o he da ase , his may no be he mos app op ia e EPP
model.
In his sec ion, addi ional in o ma ion on he Flee Age a iable (sus ainabili y da a) is
inco po a ed in he ollowing g aphs. The symbols o each piece o da a a e di e en ia ed
acco ding o he qua ile o which hey belong, consis en wi h he alue aken o ha
a iable, in acco dance wi h he symbols shown in Table 2.
Table 2.
Legend o he g aphs using he glyph me apho acco ding o he alues o he sus ainabili y-
ela ed ea u e (A e age age o he ehicle lee ).
Q Glyph
1
Sus ainabili y2022,14,xFORPEERREVIEW9o 15

HUEP o mula ionp o ideswi h hebes p ojec ion,bu oneo  henewones(KPCA)
ins ead.
Resul sIncluding heGlyphMe apho 
A e ha ingselec edKPCAas heEPPmodel ha o e s hebes p ojec ions o  he
da aanalysed, he esul susing heglyphme apho a ep esen edin hissec ion.I is
wo hmen ioning ha  o anyo he da ase , hismayno be hemos app op ia eEPP
model.
In hissec ion,addi ionalin o ma ionon heFlee Age a iable(sus ainabili yda a)
isinco po a edin he ollowingg aphs.Thesymbols o eachpieceo da aa edi e en i‐
a edacco ding o hequa ile owhich heybelong,consis en wi h he alue aken o 
ha  a iable,inacco dancewi h hesymbolsshowninTable2.
Table2.Legend o  heg aphsusing heglyphme apho acco ding o he alueso  hesus ainabil‐
i y‐ ela ed ea u e(A e ageageo  he ehicle lee ).
QGlyph
1
2
3
4

Inacco dancewi h heabo e, hep e iouslyselec edHUEPg aphisshown(Figu e
2b),al houghwi h hesus ainabili y ea u enowinco po a ed.Addi ionally, he igu eis
enhancedbya o m oguide he eade in hein e p e a iono  he esul s.
InFigu e3i ispossible osee ha aclea di e en ia iono  heda aqua ileso  he
se iesisob ained,andaclea lineo p og essioncanbema kedon he esul (dashed
yellowline).Theyellowlineisassocia ed o he empo alp og essiono da a.F oma
p ac icalpoin o  iew, heg aphshows ha  he isualisa ionob ainedisuse ulwhen
de e mining hephaseso p og essiono  heagese ies;1. heini ialage,2. hephaseo 
o e ‐amo isa iono meanso  anspo and3. hephaseo  henewa e ageageo  he
lee .

Figu e3.HUEPob ainedbyapplyingKPCAandagglome a i eclus e ing(k=3,dis ance=sEu‐
clidean,linkage=a e age) o heanalysedda ase ,using hesus ainabili y ea u ein heglyphme ‐
apho .Theyellowlineisassocia ed o he empo alp og essiono da a.
2
Sus ainabili y2022,14,xFORPEERREVIEW9o 15

HUEP o mula ionp o ideswi h hebes p ojec ion,bu oneo  henewones(KPCA)
ins ead.
Resul sIncluding heGlyphMe apho 
A e ha ingselec edKPCAas heEPPmodel ha o e s hebes p ojec ions o  he
da aanalysed, he esul susing heglyphme apho a ep esen edin hissec ion.I is
wo hmen ioning ha  o anyo he da ase , hismayno be hemos app op ia eEPP
model.
In hissec ion,addi ionalin o ma ionon heFlee Age a iable(sus ainabili yda a)
isinco po a edin he ollowingg aphs.Thesymbols o eachpieceo da aa edi e en i‐
a edacco ding o hequa ile owhich heybelong,consis en wi h he alue aken o 
ha  a iable,inacco dancewi h hesymbolsshowninTable2.
Table2.Legend o  heg aphsusing heglyphme apho acco ding o he alueso  hesus ainabil‐
i y‐ ela ed ea u e(A e ageageo  he ehicle lee ).
QGlyph
1
2
3
4

Inacco dancewi h heabo e, hep e iouslyselec edHUEPg aphisshown(Figu e
2b),al houghwi h hesus ainabili y ea u enowinco po a ed.Addi ionally, he igu eis
enhancedbya o m oguide he eade in hein e p e a iono  he esul s.
InFigu e3i ispossible osee ha aclea di e en ia iono  heda aqua ileso  he
se iesisob ained,andaclea lineo p og essioncanbema kedon he esul (dashed
yellowline).Theyellowlineisassocia ed o he empo alp og essiono da a.F oma
p ac icalpoin o  iew, heg aphshows ha  he isualisa ionob ainedisuse ulwhen
de e mining hephaseso p og essiono  heagese ies;1. heini ialage,2. hephaseo 
o e ‐amo isa iono meanso  anspo and3. hephaseo  henewa e ageageo  he
lee .

Figu e3.HUEPob ainedbyapplyingKPCAandagglome a i eclus e ing(k=3,dis ance=sEu‐
clidean,linkage=a e age) o heanalysedda ase ,using hesus ainabili y ea u ein heglyphme ‐
apho .Theyellowlineisassocia ed o he empo alp og essiono da a.
3
Sus ainabili y2022,14,xFORPEERREVIEW9o 15

HUEP o mula ionp o ideswi h hebes p ojec ion,bu oneo  henewones(KPCA)
ins ead.
Resul sIncluding heGlyphMe apho 
A e ha ingselec edKPCAas heEPPmodel ha o e s hebes p ojec ions o  he
da aanalysed, he esul susing heglyphme apho a ep esen edin hissec ion.I is
wo hmen ioning ha  o anyo he da ase , hismayno be hemos app op ia eEPP
model.
In hissec ion,addi ionalin o ma ionon heFlee Age a iable(sus ainabili yda a)
isinco po a edin he ollowingg aphs.Thesymbols o eachpieceo da aa edi e en i‐
a edacco ding o hequa ile owhich heybelong,consis en wi h he alue aken o 
ha  a iable,inacco dancewi h hesymbolsshowninTable2.
Table2.Legend o  heg aphsusing heglyphme apho acco ding o he alueso  hesus ainabil‐
i y‐ ela ed ea u e(A e ageageo  he ehicle lee ).
QGlyph
1
2
3
4

Inacco dancewi h heabo e, hep e iouslyselec edHUEPg aphisshown(Figu e
2b),al houghwi h hesus ainabili y ea u enowinco po a ed.Addi ionally, he igu eis
enhancedbya o m oguide he eade in hein e p e a iono  he esul s.
InFigu e3i ispossible osee ha aclea di e en ia iono  heda aqua ileso  he
se iesisob ained,andaclea lineo p og essioncanbema kedon he esul (dashed
yellowline).Theyellowlineisassocia ed o he empo alp og essiono da a.F oma
p ac icalpoin o  iew, heg aphshows ha  he isualisa ionob ainedisuse ulwhen
de e mining hephaseso p og essiono  heagese ies;1. heini ialage,2. hephaseo 
o e ‐amo isa iono meanso  anspo and3. hephaseo  henewa e ageageo  he
lee .

Figu e3.HUEPob ainedbyapplyingKPCAandagglome a i eclus e ing(k=3,dis ance=sEu‐
clidean,linkage=a e age) o heanalysedda ase ,using hesus ainabili y ea u ein heglyphme ‐
apho .Theyellowlineisassocia ed o he empo alp og essiono da a.
4
Sus ainabili y2022,14,xFORPEERREVIEW9o 15

HUEP o mula ionp o ideswi h hebes p ojec ion,bu oneo  henewones(KPCA)
ins ead.
Resul sIncluding heGlyphMe apho 
A e ha ingselec edKPCAas heEPPmodel ha o e s hebes p ojec ions o  he
da aanalysed, he esul susing heglyphme apho a ep esen edin hissec ion.I is
wo hmen ioning ha  o anyo he da ase , hismayno be hemos app op ia eEPP
model.
In hissec ion,addi ionalin o ma ionon heFlee Age a iable(sus ainabili yda a)
isinco po a edin he ollowingg aphs.Thesymbols o eachpieceo da aa edi e en i‐
a edacco ding o hequa ile owhich heybelong,consis en wi h he alue aken o 
ha  a iable,inacco dancewi h hesymbolsshowninTable2.
Table2.Legend o  heg aphsusing heglyphme apho acco ding o he alueso  hesus ainabil‐
i y‐ ela ed ea u e(A e ageageo  he ehicle lee ).
QGlyph
1
2
3
4

Inacco dancewi h heabo e, hep e iouslyselec edHUEPg aphisshown(Figu e
2b),al houghwi h hesus ainabili y ea u enowinco po a ed.Addi ionally, he igu eis
enhancedbya o m oguide he eade in hein e p e a iono  he esul s.
InFigu e3i ispossible osee ha aclea di e en ia iono  heda aqua ileso  he
se iesisob ained,andaclea lineo p og essioncanbema kedon he esul (dashed
yellowline).Theyellowlineisassocia ed o he empo alp og essiono da a.F oma
p ac icalpoin o  iew, heg aphshows ha  he isualisa ionob ainedisuse ulwhen
de e mining hephaseso p og essiono  heagese ies;1. heini ialage,2. hephaseo 
o e ‐amo isa iono meanso  anspo and3. hephaseo  henewa e ageageo  he
lee .

Figu e3.HUEPob ainedbyapplyingKPCAandagglome a i eclus e ing(k=3,dis ance=sEu‐
clidean,linkage=a e age) o heanalysedda ase ,using hesus ainabili y ea u ein heglyphme ‐
apho .Theyellowlineisassocia ed o he empo alp og essiono da a.
In acco dance wi h he abo e, he p e iously selec ed HUEP g aph is shown (
Figu e 2b
),
al hough wi h he sus ainabili y ea u e now inco po a ed. Addi ionally, he igu e is en-
hanced by a o m o guide he eade in he in e p e a ion o he esul s.
In Figu e 3i is possible o see ha a clea di e en ia ion o he da a qua iles o he
se ies is ob ained, and a clea line o p og ession can be ma ked on he esul (dashed
yellow line). The yellow line is associa ed o he empo al p og ession o da a. F om a
p ac ical poin o iew, he g aph shows ha he isualisa ion ob ained is use ul when
de e mining he phases o p og ession o he age se ies; 1. he ini ial age, 2. he phase
o o e -amo isa ion o means o anspo and 3. he phase o he new a e age age o
he lee .
Sus ainabili y2022,14,xFORPEERREVIEW9o 15

HUEP o mula ionp o ideswi h hebes p ojec ion,bu oneo  henewones(KPCA)
ins ead.
Resul sIncluding heGlyphMe apho 
A e ha ingselec edKPCAas heEPPmodel ha o e s hebes p ojec ions o  he
da aanalysed, he esul susing heglyphme apho a ep esen edin hissec ion.I is
wo hmen ioning ha  o anyo he da ase , hismayno be hemos app op ia eEPP
model.
In hissec ion,addi ionalin o ma ionon heFlee Age a iable(sus ainabili yda a)
isinco po a edin he ollowingg aphs.Thesymbols o eachpieceo da aa edi e en i‐
a edacco ding o hequa ile owhich heybelong,consis en wi h he alue aken o 
ha  a iable,inacco dancewi h hesymbolsshowninTable2.
Table2.Legend o  heg aphsusing heglyphme apho acco ding o he alueso  hesus ainabil‐
i y‐ ela ed ea u e(A e ageageo  he ehicle lee ).
QGlyph
1
2
3
4

Inacco dancewi h heabo e, hep e iouslyselec edHUEPg aphisshown(Figu e
2b),al houghwi h hesus ainabili y ea u enowinco po a ed.Addi ionally, he igu eis
enhancedbya o m oguide he eade in hein e p e a iono  he esul s.
InFigu e3i ispossible osee ha aclea di e en ia iono  heda aqua ileso  he
se iesisob ained,andaclea lineo p og essioncanbema kedon he esul (dashed
yellowline).Theyellowlineisassocia ed o he empo alp og essiono da a.F oma
p ac icalpoin o  iew, heg aphshows ha  he isualisa ionob ainedisuse ulwhen
de e mining hephaseso p og essiono  heagese ies;1. heini ialage,2. hephaseo 
o e ‐amo isa iono meanso  anspo and3. hephaseo  henewa e ageageo  he
lee .

Figu e3.HUEPob ainedbyapplyingKPCAandagglome a i eclus e ing(k=3,dis ance=sEu‐
clidean,linkage=a e age) o heanalysedda ase ,using hesus ainabili y ea u ein heglyphme ‐
apho .Theyellowlineisassocia ed o he empo alp og essiono da a.
Figu e 3.
HUEP ob ained by applying KPCA and agglome a i e clus e ing (k = 3, dis ance = sEu-
clidean, linkage = a e age) o he analysed da ase , using he sus ainabili y ea u e in he glyph
me apho . The yellow line is associa ed o he empo al p og ession o da a.
To app ecia e he impac o he k pa ame e o he clus e ing echnique on he display,
di e en isualisa ions a e p esen ed below (Figu e 4), wi h he same EPP and clus e ing
models, bu wi h a a ying numbe o clus e s.